The objective of our study is to evaluate the accuracy of an X-ray based image segmentation system for patient specific instrument (PSI) design or any other surgical application that requires 3D modeling of the knee.

The process requires two bilateral short film X-ray images of knee and a standing long film image of the leg including the hip and ankle. The short film images are acquired with an X-ray positioner device that is embedded with fiducial markers to correct for setup variation in source and cassette position. An automated image segmentation algorithm, based on a statistical model that couples knee bone shape and radiographic appearance, calculates 3D surface models of the knee from the bi-lateral short films (Imorphics, Manchester UK) (Figure 1). Surface silhouettes are used to inspect and refine the automatically generated segmentation; the femur and tibia mechanical axes are then calculated using automatically generated surface model landmarks combined with user-defined markups of the hip and ankle center from the standing long film (Figure 2).

The accuracy of the 2D/3D segmentation system was evaluated using simulated X-ray imagery generated from one-hundred osteoarthritic, lower limb CT image samples using the Insight Toolkit (Kitware, Inc.). Random, normally distributed variations in source and cassette positions were included in the dataset. Surface accuracy was measured using root-mean-square (RMS) point-to-surface (P2S) distance calculations with respect to paired benchmark CT segmentations. Landmark accuracy was calculated by measuring angular differences between the 2D/3D generated femur and tibia mechanical tibia with respect to paired CT-generated landmark data.

The paired RMS sample mean and standard deviation of femur P2S errors on the distal quarter of the femur after auto-segmentation was 1.08±0.20mm. The RMS sample mean and standard deviation of tibia P2S errors on the proximal quarter of the tibia after auto-segmentation was 1.16±0.25mm. The paired sample mean and standard deviation of the femur and tibia mechanical axis accuracy with respect to benchmark CT data landmarks were 0.02±0.42[deg] and −0.33±0.56[deg], respectively. Per surface-vertex sample RMS P2S errors are illustrated in Figure 3.

Visual inspection of RMS results found the automatically segmented femur to be very accurate in the shaft, distal condyles, and posterior condyles, which are important for PSI guide fit and accurate planning. Similarly, the automatically segmented tibia was very accurate in the shaft and plateaus, which are also important for PSI guide fit. Osteophytes resulted in some RMS differences (Figure 3), as was expected due to the know limitations of osteophyte imaging with X-ray. PSI-type applications that utilize X-ray should account for osteophyte segmentation error. Overall, our results based on simulated radiographic data demonstrate that X-ray based 2D/3D segmentation is a viable tool for use in orthopaedic applications that require accurate 3D segmentations of knee bones.

Introduction:

Adequate coverage of the resected tibial plateau with the tibial tray is necessary to reduce the theoretical risk of tibial subsidence after primary total knee arthroplasty (TKA). Maximizing tibial coverage is balanced against avoiding excessive overhang of the tray causing soft tissue irritation, and establishing proper tray alignment improving implant longevity and patella function¹. Implant design factors, including the number of tray sizes, tray shape, and tray asymmetry influence the ability to cover the tibial plateau². Furthermore, rotating platform (RP) tray designs decouple restoring proper tibial rotation from maximizing tibial coverage, which may enhance the ability to maximize coverage. The purpose of the current study was to assess the ability of five modern tray designs (Fig. 1), including symmetric, asymmetric, fixed-bearing, and RP designs, to maximize coverage of the tibial plateau across a large patient population.

The various disorders of the patellofemoral joint, from pain syndrome to maltracking and arthritis, form a significant subset of knee disorders (Callaghan and Selfe 2007). Several studies have shown significant geography and gender based variation in incidence rates of these disorders and of osteoarthritis in general (Woolf 2003). A number of previous studies have examined patellar shape in this context, focusing primarily on the use of 2D measurements of bony geometry to classify patellar shapes and identify high-risk groups (Baumgartl 1964; Ficat 1970).

Recent developments in imaging and statistical analysis have enabled a more sophisticated approach, characterised by statistical shape models which account for three dimensional shape differences (Bryan 2008). Incorporating soft tissue data into these analyses, however, has been a challenge due to factors including the necessity of multi-modality images, absence of repeatable landmarks, and complexity of the surfaces involved. We present here a novel method which has potential to significantly improve analysis of soft tissue geometry in joints. It is built using Arthron, a UCD-developed biomechanics analysis software package.

The shape modelling process consists of three phases: pre-processing, consistent surface parameterization, and statistical shape analysis. The pre-processing phase consists of several mesh processing operations that prepare the input surfaces for shape modelling. Consistent surface parameterizations are implemented using the minimum description length (MDL) correspondence method (Davies 2002) [Fig. 1]. The statistical shape analysis phase involves the reporting and visualization of geometric variation at the input surface. An algorithm was developed to measure the cartilage thickness at each node on the patellar surface mesh. The initial step in this process was to calculate surface normal vectors at each point. These vectors were then projected through the cartilage surface model in order to calculate the thickness [Fig. 2]. The Matlab software was used to aggregate all cartilage thickness values in a given subgroup and after being normalised for the average patellar centroid size for the subgroup, these thicknesses were visualised on the average shape.

Pilot study data consisted of 19 Caucasian (7 female, 12 male) and 13 Japanese (7 female, 6 male) subjects. These data originated from studies performed by DePuy Orthopaedics Inc. Initial results show ethnicity effects in cartilage thickness to be more significant than gender effects [Fig. 3]. After correcting for patellar centroid size, male subjects display 9% greater average thickness than female subjects, while Caucasian subjects display 17% greater average thickness than Asian subjects. Areas of statistically significant differences (t < 0.05) were found to coincide with expected areas of patellofemoral contact through the flexion cycle, showing the potential for the thickness differential to impact upon patellar kinematics. Principal component analysis of the thickness distributions gives more detailed information about modes of variation.

With further development, this method has potential to enable sophisticated analysis of localised variation in soft tissue geometry, thereby improving understanding of the impact of joint geometry on disease formation.

The management of the dysplastic hip represents a clinical and a technical challenge to the paediatric orthopaedic surgeon. There is a great deal of variation in the degree and direction of acetabular dysplasia. Preoperative planning in the dysplastic hip is still largely based on plain radiographs. However, these plain films are a 2D projection of a 3D structure and measurement is prone to inaccuracy as a result. Hip arthrography is used in an attempt to analyse the 3D morphology of the hip. However, this still employs a 2D projection of a 3D structure and in addition has the risk of general anaesthesia and infection. Geometrical analysis based on multiplanar imaging with CT scans has been shown to reduce analysis variability. We present a system for morphological analysis and preoperative of the paediatric hip using this model. Our system can be used to determine the most appropriate osteotomy based on morphology. This system should increase the accuracy of preoperative planning and reduce the need for arthrography.